arXiv 2502.155432026-07-10cs.CLcs.AI

ParamMute：抑制知识关键的前馈神经网络以实现忠实的检索增强生成

ParamMute: Suppressing Knowledge-Critical FFNs for Faithful Retrieval-Augmented Generation

School of Computer Science and Engineering, Northeastern University, China（东北大学计算机科学与工程学院）；Department of Computer Science and Technology, Institute for AI, Tsinghua University, China（清华大学人工智能研究院计算机科学与技术系）；Microsoft Research Asia, Beijing, China（微软亚洲研究院）；Language Technologies Institute, Carnegie Mellon University, United States（卡内基梅隆大学语言技术研究所）

Pengcheng Huang, Zhenghao Liu, Yukun Yan, Haiyan Zhao, Xiaoyuan Yi, Hao Chen, Zhiyuan Liu, Maosong Sun, Tong Xiao, Ge Yu, Chenyan Xiong

备注：: 26 pages, 7 figures, 7 tables

阅读 PDF

arXiv 页面 ↗HTML 原文 ↗

AI总结：

研究RAG中LLMs不忠实生成问题，提出ParamMute框架抑制相关FFNs激活并校准模型，通过CoFaithfulQA基准评估，显著提升忠实性，减少对参数记忆依赖，为提高LLM在RAG中的可信度提供新方向。

摘要：

与检索增强生成（RAG）集成的大型语言模型（LLMs）通过将输出基于外部证据来提高事实性。然而，它们仍易出现不忠实生成，即输出与检索到的上下文相矛盾。现有提高忠实性的方法主要关注增强外部上下文的利用，却常忽略生成过程中内部参数知识的持续影响。本文研究不忠实生成背后的内部机制，识别出在这种情况下过度激活的中深层前馈神经网络（FFNs）子集。基于此，提出通过FFN抑制进行参数知识静音（ParamMute）框架，通过抑制与不忠实相关的FFNs激活并使模型朝向检索知识校准来提高上下文忠实性。为评估该方法，引入CoFaithfulQA基准。实验结果表明，ParamMute在CoFaithfulQA和既定的ConFiQA基准上均显著提高了忠实性，大幅减少了对参数记忆的依赖。这些发现强调了减轻内部知识主导的重要性，并为提高RAG中LLM的可信度提供了新方向。

英文摘要：

Large language models (LLMs) integrated with retrieval-augmented generation (RAG) have improved factuality by grounding outputs in external evidence. However, they remain susceptible to unfaithful generation, where outputs contradict retrieved context despite its relevance and accuracy. Existing approaches aiming to improve faithfulness primarily focus on enhancing the utilization of external context, but often overlook the persistent influence of internal parametric knowledge during generation. In this work, we investigate the internal mechanisms behind unfaithful generation and identify a subset of mid-to-deep feed-forward networks (FFNs) that are disproportionately activated in such cases. Building on this insight, we propose Parametric Knowledge Muting through FFN Suppression (ParamMute), a framework that improves contextual faithfulness by suppressing the activation of unfaithfulness-associated FFNs and calibrating the model toward retrieved knowledge. To evaluate our approach, we introduce CoFaithfulQA, a benchmark specifically designed to evaluate faithfulness in scenarios where internal knowledge conflicts with accurate external evidence. Experimental results show that ParamMute significantly enhances faithfulness across both CoFaithfulQA and the established ConFiQA benchmark, achieving substantial reductions in reliance on parametric memory. These findings underscore the importance of mitigating internal knowledge dominance and provide a new direction for improving LLM trustworthiness in RAG. All codes are available at https://github.com/OpenBMB/ParamMute.

机构、总结、中文摘要均由AI分析而来，可能存在错误，请以论文正文为准。